On Fri, Jul 24, 2020 at 2:13 AM Kuniyuki Iwashima <kun...@amazon.co.jp> wrote:
>
> From: David Miller <da...@davemloft.net>
> Date: Thu, 23 Jul 2020 15:10:51 -0700 (PDT)
> > From: Kuniyuki Iwashima <kun...@amazon.co.jp>
> > Date: Thu, 23 Jul 2020 01:52:27 +0900
> >
> > > This patch removes an unnecessary variable in udp[46]_lib_lookup2() and
> > > makes it easier to resolve a merge conflict with bpf-next reported in
> > > the link below.
> > >
> > > Link:
> > > https://lore.kernel.org/linux-next/20200722132143.700a5...@canb.auug.org.au/
> > > Fixes: efc6b6f6c311 ("udp: Improve load balancing for SO_REUSEPORT.")
> > > Signed-off-by: Kuniyuki Iwashima <kun...@amazon.co.jp>
> > > Signed-off-by: Jakub Sitnicki <ja...@cloudflare.com>
> > > Cc: Willem de Bruijn <willemdebruijn.ker...@gmail.com>
> >
> > This doesn't apply to net-next.
>
> Yes. I think this kind of patch should be submitted to net-next, but this
> is for the net tree. Please let me add more description.
>
> Currently, the net and net-next trees conflict in udp[46]_lib_lookup2()
> between
>
> efc6b6f6c311 ("udp: Improve load balancing for SO_REUSEPORT.")
>
> and
>
> 7629c73a1466 ("udp: Extract helper for selecting socket from reuseport
> group")
> 2a08748cd384 ("udp6: Extract helper for selecting socket from reuseport
> group")
> .
>
> The conflict is reported in the link[0] and Jakub suggested how to resolve
> it[1]. To ease the merge conflict, Jakub and I have to send follow up patches
> to
> the bpf-next and net trees.
>
> Now, his patchset (7629c73a1466 and 2a08748cd384) to bpf-next is merged
> into net-next, and his follow up patch is applied in bpf-next[2].
>
> I fixed a bug in efc6b6f6c311, but it introduced an unnecessary variable
> and made the conflict worse. So I sent this follow up patch to net tree.
>
> However, I do not know the best way to resolve the conflict, so any comments
> are welcome.
Perhaps simpler is to apply this change to bpf-next:
"
badness = score;
- result = sk;
+ if (!result)
+ result = sk;
"
After which the remaining conflict between bpf-next and net is
"
++<<<<<<< HEAD
+ result = lookup_reuseport(net, sk, skb,
+ saddr, sport, daddr, hnum);
+ if (result && !reuseport_has_conns(sk, false))
+ return result;
+
+ badness = score;
+ if (!result)
+ result = sk;
++=======
+ reuseport_result = NULL;
+
+ if (sk->sk_reuseport &&
+ sk->sk_state != TCP_ESTABLISHED) {
+ hash = udp_ehashfn(net, daddr, hnum,
+ saddr, sport);
+ reuseport_result =
reuseport_select_sock(sk, hash, skb,
+
sizeof(struct udphdr));
+ if (reuseport_result &&
!reuseport_has_conns(sk, false))
+ return reuseport_result;
+ }
+
+ result = reuseport_result ? : sk;
+ badness = score;
++>>>>>>> netdev-net/master
"
And we can just take bpf-next HEAD. Either that or handle the
corresponding change in the merge fix-up itself.
(note that bpf-next is one patch ahead of netdev-nn)
